mthca_main.c revision 80fd8238734c852a8ed1ea39f8444a2df33bd161
1/*
2 * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses.  You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 *     Redistribution and use in source and binary forms, with or
12 *     without modification, are permitted provided that the following
13 *     conditions are met:
14 *
15 *      - Redistributions of source code must retain the above
16 *        copyright notice, this list of conditions and the following
17 *        disclaimer.
18 *
19 *      - Redistributions in binary form must reproduce the above
20 *        copyright notice, this list of conditions and the following
21 *        disclaimer in the documentation and/or other materials
22 *        provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 *
33 * $Id: mthca_main.c 1396 2004-12-28 04:10:27Z roland $
34 */
35
36#include <linux/config.h>
37#include <linux/version.h>
38#include <linux/module.h>
39#include <linux/init.h>
40#include <linux/errno.h>
41#include <linux/pci.h>
42#include <linux/interrupt.h>
43
44#include "mthca_dev.h"
45#include "mthca_config_reg.h"
46#include "mthca_cmd.h"
47#include "mthca_profile.h"
48#include "mthca_memfree.h"
49
50MODULE_AUTHOR("Roland Dreier");
51MODULE_DESCRIPTION("Mellanox InfiniBand HCA low-level driver");
52MODULE_LICENSE("Dual BSD/GPL");
53MODULE_VERSION(DRV_VERSION);
54
55#ifdef CONFIG_PCI_MSI
56
57static int msi_x = 0;
58module_param(msi_x, int, 0444);
59MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
60
61static int msi = 0;
62module_param(msi, int, 0444);
63MODULE_PARM_DESC(msi, "attempt to use MSI if nonzero");
64
65#else /* CONFIG_PCI_MSI */
66
67#define msi_x (0)
68#define msi   (0)
69
70#endif /* CONFIG_PCI_MSI */
71
72static const char mthca_version[] __devinitdata =
73	DRV_NAME ": Mellanox InfiniBand HCA driver v"
74	DRV_VERSION " (" DRV_RELDATE ")\n";
75
76static struct mthca_profile default_profile = {
77	.num_qp		   = 1 << 16,
78	.rdb_per_qp	   = 4,
79	.num_cq		   = 1 << 16,
80	.num_mcg	   = 1 << 13,
81	.num_mpt	   = 1 << 17,
82	.num_mtt	   = 1 << 20,
83	.num_udav	   = 1 << 15,	/* Tavor only */
84	.fmr_reserved_mtts = 1 << 18,	/* Tavor only */
85	.uarc_size	   = 1 << 18,	/* Arbel only */
86};
87
88static int __devinit mthca_tune_pci(struct mthca_dev *mdev)
89{
90	int cap;
91	u16 val;
92
93	/* First try to max out Read Byte Count */
94	cap = pci_find_capability(mdev->pdev, PCI_CAP_ID_PCIX);
95	if (cap) {
96		if (pci_read_config_word(mdev->pdev, cap + PCI_X_CMD, &val)) {
97			mthca_err(mdev, "Couldn't read PCI-X command register, "
98				  "aborting.\n");
99			return -ENODEV;
100		}
101		val = (val & ~PCI_X_CMD_MAX_READ) | (3 << 2);
102		if (pci_write_config_word(mdev->pdev, cap + PCI_X_CMD, val)) {
103			mthca_err(mdev, "Couldn't write PCI-X command register, "
104				  "aborting.\n");
105			return -ENODEV;
106		}
107	} else if (!(mdev->mthca_flags & MTHCA_FLAG_PCIE))
108		mthca_info(mdev, "No PCI-X capability, not setting RBC.\n");
109
110	cap = pci_find_capability(mdev->pdev, PCI_CAP_ID_EXP);
111	if (cap) {
112		if (pci_read_config_word(mdev->pdev, cap + PCI_EXP_DEVCTL, &val)) {
113			mthca_err(mdev, "Couldn't read PCI Express device control "
114				  "register, aborting.\n");
115			return -ENODEV;
116		}
117		val = (val & ~PCI_EXP_DEVCTL_READRQ) | (5 << 12);
118		if (pci_write_config_word(mdev->pdev, cap + PCI_EXP_DEVCTL, val)) {
119			mthca_err(mdev, "Couldn't write PCI Express device control "
120				  "register, aborting.\n");
121			return -ENODEV;
122		}
123	} else if (mdev->mthca_flags & MTHCA_FLAG_PCIE)
124		mthca_info(mdev, "No PCI Express capability, "
125			   "not setting Max Read Request Size.\n");
126
127	return 0;
128}
129
130static int __devinit mthca_dev_lim(struct mthca_dev *mdev, struct mthca_dev_lim *dev_lim)
131{
132	int err;
133	u8 status;
134
135	err = mthca_QUERY_DEV_LIM(mdev, dev_lim, &status);
136	if (err) {
137		mthca_err(mdev, "QUERY_DEV_LIM command failed, aborting.\n");
138		return err;
139	}
140	if (status) {
141		mthca_err(mdev, "QUERY_DEV_LIM returned status 0x%02x, "
142			  "aborting.\n", status);
143		return -EINVAL;
144	}
145	if (dev_lim->min_page_sz > PAGE_SIZE) {
146		mthca_err(mdev, "HCA minimum page size of %d bigger than "
147			  "kernel PAGE_SIZE of %ld, aborting.\n",
148			  dev_lim->min_page_sz, PAGE_SIZE);
149		return -ENODEV;
150	}
151	if (dev_lim->num_ports > MTHCA_MAX_PORTS) {
152		mthca_err(mdev, "HCA has %d ports, but we only support %d, "
153			  "aborting.\n",
154			  dev_lim->num_ports, MTHCA_MAX_PORTS);
155		return -ENODEV;
156	}
157
158	mdev->limits.num_ports      	= dev_lim->num_ports;
159	mdev->limits.vl_cap             = dev_lim->max_vl;
160	mdev->limits.mtu_cap            = dev_lim->max_mtu;
161	mdev->limits.gid_table_len  	= dev_lim->max_gids;
162	mdev->limits.pkey_table_len 	= dev_lim->max_pkeys;
163	mdev->limits.local_ca_ack_delay = dev_lim->local_ca_ack_delay;
164	mdev->limits.max_sg             = dev_lim->max_sg;
165	mdev->limits.reserved_qps       = dev_lim->reserved_qps;
166	mdev->limits.reserved_srqs      = dev_lim->reserved_srqs;
167	mdev->limits.reserved_eecs      = dev_lim->reserved_eecs;
168	mdev->limits.reserved_cqs       = dev_lim->reserved_cqs;
169	mdev->limits.reserved_eqs       = dev_lim->reserved_eqs;
170	mdev->limits.reserved_mtts      = dev_lim->reserved_mtts;
171	mdev->limits.reserved_mrws      = dev_lim->reserved_mrws;
172	mdev->limits.reserved_uars      = dev_lim->reserved_uars;
173	mdev->limits.reserved_pds       = dev_lim->reserved_pds;
174
175	/* IB_DEVICE_RESIZE_MAX_WR not supported by driver.
176	   May be doable since hardware supports it for SRQ.
177
178	   IB_DEVICE_N_NOTIFY_CQ is supported by hardware but not by driver.
179
180	   IB_DEVICE_SRQ_RESIZE is supported by hardware but SRQ is not
181	   supported by driver. */
182	mdev->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
183		IB_DEVICE_PORT_ACTIVE_EVENT |
184		IB_DEVICE_SYS_IMAGE_GUID |
185		IB_DEVICE_RC_RNR_NAK_GEN;
186
187	if (dev_lim->flags & DEV_LIM_FLAG_BAD_PKEY_CNTR)
188		mdev->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
189
190	if (dev_lim->flags & DEV_LIM_FLAG_BAD_QKEY_CNTR)
191		mdev->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
192
193	if (dev_lim->flags & DEV_LIM_FLAG_RAW_MULTI)
194		mdev->device_cap_flags |= IB_DEVICE_RAW_MULTI;
195
196	if (dev_lim->flags & DEV_LIM_FLAG_AUTO_PATH_MIG)
197		mdev->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
198
199	if (dev_lim->flags & DEV_LIM_FLAG_UD_AV_PORT_ENFORCE)
200		mdev->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
201
202	if (dev_lim->flags & DEV_LIM_FLAG_SRQ)
203		mdev->mthca_flags |= MTHCA_FLAG_SRQ;
204
205	return 0;
206}
207
208static int __devinit mthca_init_tavor(struct mthca_dev *mdev)
209{
210	u8 status;
211	int err;
212	struct mthca_dev_lim        dev_lim;
213	struct mthca_profile        profile;
214	struct mthca_init_hca_param init_hca;
215	struct mthca_adapter        adapter;
216
217	err = mthca_SYS_EN(mdev, &status);
218	if (err) {
219		mthca_err(mdev, "SYS_EN command failed, aborting.\n");
220		return err;
221	}
222	if (status) {
223		mthca_err(mdev, "SYS_EN returned status 0x%02x, "
224			  "aborting.\n", status);
225		return -EINVAL;
226	}
227
228	err = mthca_QUERY_FW(mdev, &status);
229	if (err) {
230		mthca_err(mdev, "QUERY_FW command failed, aborting.\n");
231		goto err_disable;
232	}
233	if (status) {
234		mthca_err(mdev, "QUERY_FW returned status 0x%02x, "
235			  "aborting.\n", status);
236		err = -EINVAL;
237		goto err_disable;
238	}
239	err = mthca_QUERY_DDR(mdev, &status);
240	if (err) {
241		mthca_err(mdev, "QUERY_DDR command failed, aborting.\n");
242		goto err_disable;
243	}
244	if (status) {
245		mthca_err(mdev, "QUERY_DDR returned status 0x%02x, "
246			  "aborting.\n", status);
247		err = -EINVAL;
248		goto err_disable;
249	}
250
251	err = mthca_dev_lim(mdev, &dev_lim);
252
253	profile = default_profile;
254	profile.num_uar   = dev_lim.uar_size / PAGE_SIZE;
255	profile.uarc_size = 0;
256
257	err = mthca_make_profile(mdev, &profile, &dev_lim, &init_hca);
258	if (err < 0)
259		goto err_disable;
260
261	err = mthca_INIT_HCA(mdev, &init_hca, &status);
262	if (err) {
263		mthca_err(mdev, "INIT_HCA command failed, aborting.\n");
264		goto err_disable;
265	}
266	if (status) {
267		mthca_err(mdev, "INIT_HCA returned status 0x%02x, "
268			  "aborting.\n", status);
269		err = -EINVAL;
270		goto err_disable;
271	}
272
273	err = mthca_QUERY_ADAPTER(mdev, &adapter, &status);
274	if (err) {
275		mthca_err(mdev, "QUERY_ADAPTER command failed, aborting.\n");
276		goto err_close;
277	}
278	if (status) {
279		mthca_err(mdev, "QUERY_ADAPTER returned status 0x%02x, "
280			  "aborting.\n", status);
281		err = -EINVAL;
282		goto err_close;
283	}
284
285	mdev->eq_table.inta_pin = adapter.inta_pin;
286	mdev->rev_id            = adapter.revision_id;
287
288	return 0;
289
290err_close:
291	mthca_CLOSE_HCA(mdev, 0, &status);
292
293err_disable:
294	mthca_SYS_DIS(mdev, &status);
295
296	return err;
297}
298
299static int __devinit mthca_load_fw(struct mthca_dev *mdev)
300{
301	u8 status;
302	int err;
303
304	/* FIXME: use HCA-attached memory for FW if present */
305
306	mdev->fw.arbel.fw_icm =
307		mthca_alloc_icm(mdev, mdev->fw.arbel.fw_pages,
308				GFP_HIGHUSER | __GFP_NOWARN);
309	if (!mdev->fw.arbel.fw_icm) {
310		mthca_err(mdev, "Couldn't allocate FW area, aborting.\n");
311		return -ENOMEM;
312	}
313
314	err = mthca_MAP_FA(mdev, mdev->fw.arbel.fw_icm, &status);
315	if (err) {
316		mthca_err(mdev, "MAP_FA command failed, aborting.\n");
317		goto err_free;
318	}
319	if (status) {
320		mthca_err(mdev, "MAP_FA returned status 0x%02x, aborting.\n", status);
321		err = -EINVAL;
322		goto err_free;
323	}
324	err = mthca_RUN_FW(mdev, &status);
325	if (err) {
326		mthca_err(mdev, "RUN_FW command failed, aborting.\n");
327		goto err_unmap_fa;
328	}
329	if (status) {
330		mthca_err(mdev, "RUN_FW returned status 0x%02x, aborting.\n", status);
331		err = -EINVAL;
332		goto err_unmap_fa;
333	}
334
335	return 0;
336
337err_unmap_fa:
338	mthca_UNMAP_FA(mdev, &status);
339
340err_free:
341	mthca_free_icm(mdev, mdev->fw.arbel.fw_icm);
342	return err;
343}
344
345static int __devinit mthca_init_icm(struct mthca_dev *mdev,
346				    struct mthca_dev_lim *dev_lim,
347				    struct mthca_init_hca_param *init_hca,
348				    u64 icm_size)
349{
350	u64 aux_pages;
351	u8 status;
352	int err;
353
354	err = mthca_SET_ICM_SIZE(mdev, icm_size, &aux_pages, &status);
355	if (err) {
356		mthca_err(mdev, "SET_ICM_SIZE command failed, aborting.\n");
357		return err;
358	}
359	if (status) {
360		mthca_err(mdev, "SET_ICM_SIZE returned status 0x%02x, "
361			  "aborting.\n", status);
362		return -EINVAL;
363	}
364
365	mthca_dbg(mdev, "%lld KB of HCA context requires %lld KB aux memory.\n",
366		  (unsigned long long) icm_size >> 10,
367		  (unsigned long long) aux_pages << 2);
368
369	mdev->fw.arbel.aux_icm = mthca_alloc_icm(mdev, aux_pages,
370						 GFP_HIGHUSER | __GFP_NOWARN);
371	if (!mdev->fw.arbel.aux_icm) {
372		mthca_err(mdev, "Couldn't allocate aux memory, aborting.\n");
373		return -ENOMEM;
374	}
375
376	err = mthca_MAP_ICM_AUX(mdev, mdev->fw.arbel.aux_icm, &status);
377	if (err) {
378		mthca_err(mdev, "MAP_ICM_AUX command failed, aborting.\n");
379		goto err_free_aux;
380	}
381	if (status) {
382		mthca_err(mdev, "MAP_ICM_AUX returned status 0x%02x, aborting.\n", status);
383		err = -EINVAL;
384		goto err_free_aux;
385	}
386
387	err = mthca_map_eq_icm(mdev, init_hca->eqc_base);
388	if (err) {
389		mthca_err(mdev, "Failed to map EQ context memory, aborting.\n");
390		goto err_unmap_aux;
391	}
392
393	mdev->mr_table.mtt_table = mthca_alloc_icm_table(mdev, init_hca->mtt_base,
394							 MTHCA_MTT_SEG_SIZE,
395							 mdev->limits.num_mtt_segs,
396							 mdev->limits.reserved_mtts, 1);
397	if (!mdev->mr_table.mtt_table) {
398		mthca_err(mdev, "Failed to map MTT context memory, aborting.\n");
399		err = -ENOMEM;
400		goto err_unmap_eq;
401	}
402
403	mdev->mr_table.mpt_table = mthca_alloc_icm_table(mdev, init_hca->mpt_base,
404							 dev_lim->mpt_entry_sz,
405							 mdev->limits.num_mpts,
406							 mdev->limits.reserved_mrws, 1);
407	if (!mdev->mr_table.mpt_table) {
408		mthca_err(mdev, "Failed to map MPT context memory, aborting.\n");
409		err = -ENOMEM;
410		goto err_unmap_mtt;
411	}
412
413	mdev->qp_table.qp_table = mthca_alloc_icm_table(mdev, init_hca->qpc_base,
414							dev_lim->qpc_entry_sz,
415							mdev->limits.num_qps,
416							mdev->limits.reserved_qps, 0);
417	if (!mdev->qp_table.qp_table) {
418		mthca_err(mdev, "Failed to map QP context memory, aborting.\n");
419		err = -ENOMEM;
420		goto err_unmap_mpt;
421	}
422
423	mdev->qp_table.eqp_table = mthca_alloc_icm_table(mdev, init_hca->eqpc_base,
424							 dev_lim->eqpc_entry_sz,
425							 mdev->limits.num_qps,
426							 mdev->limits.reserved_qps, 0);
427	if (!mdev->qp_table.eqp_table) {
428		mthca_err(mdev, "Failed to map EQP context memory, aborting.\n");
429		err = -ENOMEM;
430		goto err_unmap_qp;
431	}
432
433	mdev->qp_table.rdb_table = mthca_alloc_icm_table(mdev, init_hca->rdb_base,
434							 MTHCA_RDB_ENTRY_SIZE,
435							 mdev->limits.num_qps <<
436							 mdev->qp_table.rdb_shift,
437							 0, 0);
438	if (!mdev->qp_table.rdb_table) {
439		mthca_err(mdev, "Failed to map RDB context memory, aborting\n");
440		err = -ENOMEM;
441		goto err_unmap_eqp;
442	}
443
444       mdev->cq_table.table = mthca_alloc_icm_table(mdev, init_hca->cqc_base,
445						     dev_lim->cqc_entry_sz,
446						     mdev->limits.num_cqs,
447						     mdev->limits.reserved_cqs, 0);
448	if (!mdev->cq_table.table) {
449		mthca_err(mdev, "Failed to map CQ context memory, aborting.\n");
450		err = -ENOMEM;
451		goto err_unmap_rdb;
452	}
453
454	/*
455	 * It's not strictly required, but for simplicity just map the
456	 * whole multicast group table now.  The table isn't very big
457	 * and it's a lot easier than trying to track ref counts.
458	 */
459	mdev->mcg_table.table = mthca_alloc_icm_table(mdev, init_hca->mc_base,
460						      MTHCA_MGM_ENTRY_SIZE,
461						      mdev->limits.num_mgms +
462						      mdev->limits.num_amgms,
463						      mdev->limits.num_mgms +
464						      mdev->limits.num_amgms,
465						      0);
466	if (!mdev->mcg_table.table) {
467		mthca_err(mdev, "Failed to map MCG context memory, aborting.\n");
468		err = -ENOMEM;
469		goto err_unmap_cq;
470	}
471
472	return 0;
473
474err_unmap_cq:
475	mthca_free_icm_table(mdev, mdev->cq_table.table);
476
477err_unmap_rdb:
478	mthca_free_icm_table(mdev, mdev->qp_table.rdb_table);
479
480err_unmap_eqp:
481	mthca_free_icm_table(mdev, mdev->qp_table.eqp_table);
482
483err_unmap_qp:
484	mthca_free_icm_table(mdev, mdev->qp_table.qp_table);
485
486err_unmap_mpt:
487	mthca_free_icm_table(mdev, mdev->mr_table.mpt_table);
488
489err_unmap_mtt:
490	mthca_free_icm_table(mdev, mdev->mr_table.mtt_table);
491
492err_unmap_eq:
493	mthca_unmap_eq_icm(mdev);
494
495err_unmap_aux:
496	mthca_UNMAP_ICM_AUX(mdev, &status);
497
498err_free_aux:
499	mthca_free_icm(mdev, mdev->fw.arbel.aux_icm);
500
501	return err;
502}
503
504static int __devinit mthca_init_arbel(struct mthca_dev *mdev)
505{
506	struct mthca_dev_lim        dev_lim;
507	struct mthca_profile        profile;
508	struct mthca_init_hca_param init_hca;
509	struct mthca_adapter        adapter;
510	u64 icm_size;
511	u8 status;
512	int err;
513
514	err = mthca_QUERY_FW(mdev, &status);
515	if (err) {
516		mthca_err(mdev, "QUERY_FW command failed, aborting.\n");
517		return err;
518	}
519	if (status) {
520		mthca_err(mdev, "QUERY_FW returned status 0x%02x, "
521			  "aborting.\n", status);
522		return -EINVAL;
523	}
524
525	err = mthca_ENABLE_LAM(mdev, &status);
526	if (err) {
527		mthca_err(mdev, "ENABLE_LAM command failed, aborting.\n");
528		return err;
529	}
530	if (status == MTHCA_CMD_STAT_LAM_NOT_PRE) {
531		mthca_dbg(mdev, "No HCA-attached memory (running in MemFree mode)\n");
532		mdev->mthca_flags |= MTHCA_FLAG_NO_LAM;
533	} else if (status) {
534		mthca_err(mdev, "ENABLE_LAM returned status 0x%02x, "
535			  "aborting.\n", status);
536		return -EINVAL;
537	}
538
539	err = mthca_load_fw(mdev);
540	if (err) {
541		mthca_err(mdev, "Failed to start FW, aborting.\n");
542		goto err_disable;
543	}
544
545	err = mthca_dev_lim(mdev, &dev_lim);
546	if (err) {
547		mthca_err(mdev, "QUERY_DEV_LIM command failed, aborting.\n");
548		goto err_stop_fw;
549	}
550
551	profile = default_profile;
552	profile.num_uar  = dev_lim.uar_size / PAGE_SIZE;
553	profile.num_udav = 0;
554
555	icm_size = mthca_make_profile(mdev, &profile, &dev_lim, &init_hca);
556	if ((int) icm_size < 0) {
557		err = icm_size;
558		goto err_stop_fw;
559	}
560
561	err = mthca_init_icm(mdev, &dev_lim, &init_hca, icm_size);
562	if (err)
563		goto err_stop_fw;
564
565	err = mthca_INIT_HCA(mdev, &init_hca, &status);
566	if (err) {
567		mthca_err(mdev, "INIT_HCA command failed, aborting.\n");
568		goto err_free_icm;
569	}
570	if (status) {
571		mthca_err(mdev, "INIT_HCA returned status 0x%02x, "
572			  "aborting.\n", status);
573		err = -EINVAL;
574		goto err_free_icm;
575	}
576
577	err = mthca_QUERY_ADAPTER(mdev, &adapter, &status);
578	if (err) {
579		mthca_err(mdev, "QUERY_ADAPTER command failed, aborting.\n");
580		goto err_free_icm;
581	}
582	if (status) {
583		mthca_err(mdev, "QUERY_ADAPTER returned status 0x%02x, "
584			  "aborting.\n", status);
585		err = -EINVAL;
586		goto err_free_icm;
587	}
588
589	mdev->eq_table.inta_pin = adapter.inta_pin;
590	mdev->rev_id            = adapter.revision_id;
591
592	return 0;
593
594err_free_icm:
595	mthca_free_icm_table(mdev, mdev->cq_table.table);
596	mthca_free_icm_table(mdev, mdev->qp_table.rdb_table);
597	mthca_free_icm_table(mdev, mdev->qp_table.eqp_table);
598	mthca_free_icm_table(mdev, mdev->qp_table.qp_table);
599	mthca_free_icm_table(mdev, mdev->mr_table.mpt_table);
600	mthca_free_icm_table(mdev, mdev->mr_table.mtt_table);
601	mthca_unmap_eq_icm(mdev);
602
603	mthca_UNMAP_ICM_AUX(mdev, &status);
604	mthca_free_icm(mdev, mdev->fw.arbel.aux_icm);
605
606err_stop_fw:
607	mthca_UNMAP_FA(mdev, &status);
608	mthca_free_icm(mdev, mdev->fw.arbel.fw_icm);
609
610err_disable:
611	if (!(mdev->mthca_flags & MTHCA_FLAG_NO_LAM))
612		mthca_DISABLE_LAM(mdev, &status);
613
614	return err;
615}
616
617static int __devinit mthca_init_hca(struct mthca_dev *mdev)
618{
619	if (mthca_is_memfree(mdev))
620		return mthca_init_arbel(mdev);
621	else
622		return mthca_init_tavor(mdev);
623}
624
625static int __devinit mthca_setup_hca(struct mthca_dev *dev)
626{
627	int err;
628	u8 status;
629
630	MTHCA_INIT_DOORBELL_LOCK(&dev->doorbell_lock);
631
632	err = mthca_init_uar_table(dev);
633	if (err) {
634		mthca_err(dev, "Failed to initialize "
635			  "user access region table, aborting.\n");
636		return err;
637	}
638
639	err = mthca_uar_alloc(dev, &dev->driver_uar);
640	if (err) {
641		mthca_err(dev, "Failed to allocate driver access region, "
642			  "aborting.\n");
643		goto err_uar_table_free;
644	}
645
646	dev->kar = ioremap(dev->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
647	if (!dev->kar) {
648		mthca_err(dev, "Couldn't map kernel access region, "
649			  "aborting.\n");
650		err = -ENOMEM;
651		goto err_uar_free;
652	}
653
654	err = mthca_init_pd_table(dev);
655	if (err) {
656		mthca_err(dev, "Failed to initialize "
657			  "protection domain table, aborting.\n");
658		goto err_kar_unmap;
659	}
660
661	err = mthca_init_mr_table(dev);
662	if (err) {
663		mthca_err(dev, "Failed to initialize "
664			  "memory region table, aborting.\n");
665		goto err_pd_table_free;
666	}
667
668	err = mthca_pd_alloc(dev, &dev->driver_pd);
669	if (err) {
670		mthca_err(dev, "Failed to create driver PD, "
671			  "aborting.\n");
672		goto err_mr_table_free;
673	}
674
675	err = mthca_init_eq_table(dev);
676	if (err) {
677		mthca_err(dev, "Failed to initialize "
678			  "event queue table, aborting.\n");
679		goto err_pd_free;
680	}
681
682	err = mthca_cmd_use_events(dev);
683	if (err) {
684		mthca_err(dev, "Failed to switch to event-driven "
685			  "firmware commands, aborting.\n");
686		goto err_eq_table_free;
687	}
688
689	err = mthca_NOP(dev, &status);
690	if (err || status) {
691		mthca_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting.\n",
692			  dev->mthca_flags & MTHCA_FLAG_MSI_X ?
693			  dev->eq_table.eq[MTHCA_EQ_CMD].msi_x_vector :
694			  dev->pdev->irq);
695		if (dev->mthca_flags & (MTHCA_FLAG_MSI | MTHCA_FLAG_MSI_X))
696			mthca_err(dev, "Try again with MSI/MSI-X disabled.\n");
697		else
698			mthca_err(dev, "BIOS or ACPI interrupt routing problem?\n");
699
700		goto err_cmd_poll;
701	}
702
703	mthca_dbg(dev, "NOP command IRQ test passed\n");
704
705	err = mthca_init_cq_table(dev);
706	if (err) {
707		mthca_err(dev, "Failed to initialize "
708			  "completion queue table, aborting.\n");
709		goto err_cmd_poll;
710	}
711
712	err = mthca_init_qp_table(dev);
713	if (err) {
714		mthca_err(dev, "Failed to initialize "
715			  "queue pair table, aborting.\n");
716		goto err_cq_table_free;
717	}
718
719	err = mthca_init_av_table(dev);
720	if (err) {
721		mthca_err(dev, "Failed to initialize "
722			  "address vector table, aborting.\n");
723		goto err_qp_table_free;
724	}
725
726	err = mthca_init_mcg_table(dev);
727	if (err) {
728		mthca_err(dev, "Failed to initialize "
729			  "multicast group table, aborting.\n");
730		goto err_av_table_free;
731	}
732
733	return 0;
734
735err_av_table_free:
736	mthca_cleanup_av_table(dev);
737
738err_qp_table_free:
739	mthca_cleanup_qp_table(dev);
740
741err_cq_table_free:
742	mthca_cleanup_cq_table(dev);
743
744err_cmd_poll:
745	mthca_cmd_use_polling(dev);
746
747err_eq_table_free:
748	mthca_cleanup_eq_table(dev);
749
750err_pd_free:
751	mthca_pd_free(dev, &dev->driver_pd);
752
753err_mr_table_free:
754	mthca_cleanup_mr_table(dev);
755
756err_pd_table_free:
757	mthca_cleanup_pd_table(dev);
758
759err_kar_unmap:
760	iounmap(dev->kar);
761
762err_uar_free:
763	mthca_uar_free(dev, &dev->driver_uar);
764
765err_uar_table_free:
766	mthca_cleanup_uar_table(dev);
767	return err;
768}
769
770static int __devinit mthca_request_regions(struct pci_dev *pdev,
771					   int ddr_hidden)
772{
773	int err;
774
775	/*
776	 * We can't just use pci_request_regions() because the MSI-X
777	 * table is right in the middle of the first BAR.  If we did
778	 * pci_request_region and grab all of the first BAR, then
779	 * setting up MSI-X would fail, since the PCI core wants to do
780	 * request_mem_region on the MSI-X vector table.
781	 *
782	 * So just request what we need right now, and request any
783	 * other regions we need when setting up EQs.
784	 */
785	if (!request_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
786				MTHCA_HCR_SIZE, DRV_NAME))
787		return -EBUSY;
788
789	err = pci_request_region(pdev, 2, DRV_NAME);
790	if (err)
791		goto err_bar2_failed;
792
793	if (!ddr_hidden) {
794		err = pci_request_region(pdev, 4, DRV_NAME);
795		if (err)
796			goto err_bar4_failed;
797	}
798
799	return 0;
800
801err_bar4_failed:
802	pci_release_region(pdev, 2);
803
804err_bar2_failed:
805	release_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
806			   MTHCA_HCR_SIZE);
807
808	return err;
809}
810
811static void mthca_release_regions(struct pci_dev *pdev,
812				  int ddr_hidden)
813{
814	if (!ddr_hidden)
815		pci_release_region(pdev, 4);
816
817	pci_release_region(pdev, 2);
818
819	release_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
820			   MTHCA_HCR_SIZE);
821}
822
823static int __devinit mthca_enable_msi_x(struct mthca_dev *mdev)
824{
825	struct msix_entry entries[3];
826	int err;
827
828	entries[0].entry = 0;
829	entries[1].entry = 1;
830	entries[2].entry = 2;
831
832	err = pci_enable_msix(mdev->pdev, entries, ARRAY_SIZE(entries));
833	if (err) {
834		if (err > 0)
835			mthca_info(mdev, "Only %d MSI-X vectors available, "
836				   "not using MSI-X\n", err);
837		return err;
838	}
839
840	mdev->eq_table.eq[MTHCA_EQ_COMP ].msi_x_vector = entries[0].vector;
841	mdev->eq_table.eq[MTHCA_EQ_ASYNC].msi_x_vector = entries[1].vector;
842	mdev->eq_table.eq[MTHCA_EQ_CMD  ].msi_x_vector = entries[2].vector;
843
844	return 0;
845}
846
847static void mthca_close_hca(struct mthca_dev *mdev)
848{
849	u8 status;
850
851	mthca_CLOSE_HCA(mdev, 0, &status);
852
853	if (mthca_is_memfree(mdev)) {
854		mthca_free_icm_table(mdev, mdev->cq_table.table);
855		mthca_free_icm_table(mdev, mdev->qp_table.rdb_table);
856		mthca_free_icm_table(mdev, mdev->qp_table.eqp_table);
857		mthca_free_icm_table(mdev, mdev->qp_table.qp_table);
858		mthca_free_icm_table(mdev, mdev->mr_table.mpt_table);
859		mthca_free_icm_table(mdev, mdev->mr_table.mtt_table);
860		mthca_unmap_eq_icm(mdev);
861
862		mthca_UNMAP_ICM_AUX(mdev, &status);
863		mthca_free_icm(mdev, mdev->fw.arbel.aux_icm);
864
865		mthca_UNMAP_FA(mdev, &status);
866		mthca_free_icm(mdev, mdev->fw.arbel.fw_icm);
867
868		if (!(mdev->mthca_flags & MTHCA_FLAG_NO_LAM))
869			mthca_DISABLE_LAM(mdev, &status);
870	} else
871		mthca_SYS_DIS(mdev, &status);
872}
873
874/* Types of supported HCA */
875enum {
876	TAVOR,			/* MT23108                        */
877	ARBEL_COMPAT,		/* MT25208 in Tavor compat mode   */
878	ARBEL_NATIVE,		/* MT25208 with extended features */
879	SINAI			/* MT25204 */
880};
881
882#define MTHCA_FW_VER(major, minor, subminor) \
883	(((u64) (major) << 32) | ((u64) (minor) << 16) | (u64) (subminor))
884
885static struct {
886	u64 latest_fw;
887	int is_memfree;
888	int is_pcie;
889} mthca_hca_table[] = {
890	[TAVOR]        = { .latest_fw = MTHCA_FW_VER(3, 3, 2), .is_memfree = 0, .is_pcie = 0 },
891	[ARBEL_COMPAT] = { .latest_fw = MTHCA_FW_VER(4, 6, 2), .is_memfree = 0, .is_pcie = 1 },
892	[ARBEL_NATIVE] = { .latest_fw = MTHCA_FW_VER(5, 0, 1), .is_memfree = 1, .is_pcie = 1 },
893	[SINAI]        = { .latest_fw = MTHCA_FW_VER(1, 0, 1), .is_memfree = 1, .is_pcie = 1 }
894};
895
896static int __devinit mthca_init_one(struct pci_dev *pdev,
897				    const struct pci_device_id *id)
898{
899	static int mthca_version_printed = 0;
900	int ddr_hidden = 0;
901	int err;
902	struct mthca_dev *mdev;
903
904	if (!mthca_version_printed) {
905		printk(KERN_INFO "%s", mthca_version);
906		++mthca_version_printed;
907	}
908
909	printk(KERN_INFO PFX "Initializing %s (%s)\n",
910	       pci_pretty_name(pdev), pci_name(pdev));
911
912	if (id->driver_data >= ARRAY_SIZE(mthca_hca_table)) {
913		printk(KERN_ERR PFX "%s (%s) has invalid driver data %lx\n",
914		       pci_pretty_name(pdev), pci_name(pdev), id->driver_data);
915		return -ENODEV;
916	}
917
918	err = pci_enable_device(pdev);
919	if (err) {
920		dev_err(&pdev->dev, "Cannot enable PCI device, "
921			"aborting.\n");
922		return err;
923	}
924
925	/*
926	 * Check for BARs.  We expect 0: 1MB, 2: 8MB, 4: DDR (may not
927	 * be present)
928	 */
929	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
930	    pci_resource_len(pdev, 0) != 1 << 20) {
931		dev_err(&pdev->dev, "Missing DCS, aborting.\n");
932		err = -ENODEV;
933		goto err_disable_pdev;
934	}
935	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM) ||
936	    pci_resource_len(pdev, 2) != 1 << 23) {
937		dev_err(&pdev->dev, "Missing UAR, aborting.\n");
938		err = -ENODEV;
939		goto err_disable_pdev;
940	}
941	if (!(pci_resource_flags(pdev, 4) & IORESOURCE_MEM))
942		ddr_hidden = 1;
943
944	err = mthca_request_regions(pdev, ddr_hidden);
945	if (err) {
946		dev_err(&pdev->dev, "Cannot obtain PCI resources, "
947			"aborting.\n");
948		goto err_disable_pdev;
949	}
950
951	pci_set_master(pdev);
952
953	err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
954	if (err) {
955		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
956		err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
957		if (err) {
958			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
959			goto err_free_res;
960		}
961	}
962	err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
963	if (err) {
964		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
965			 "consistent PCI DMA mask.\n");
966		err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
967		if (err) {
968			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
969				"aborting.\n");
970			goto err_free_res;
971		}
972	}
973
974	mdev = (struct mthca_dev *) ib_alloc_device(sizeof *mdev);
975	if (!mdev) {
976		dev_err(&pdev->dev, "Device struct alloc failed, "
977			"aborting.\n");
978		err = -ENOMEM;
979		goto err_free_res;
980	}
981
982	mdev->pdev = pdev;
983
984	if (ddr_hidden)
985		mdev->mthca_flags |= MTHCA_FLAG_DDR_HIDDEN;
986	if (mthca_hca_table[id->driver_data].is_memfree)
987		mdev->mthca_flags |= MTHCA_FLAG_MEMFREE;
988	if (mthca_hca_table[id->driver_data].is_pcie)
989		mdev->mthca_flags |= MTHCA_FLAG_PCIE;
990
991	/*
992	 * Now reset the HCA before we touch the PCI capabilities or
993	 * attempt a firmware command, since a boot ROM may have left
994	 * the HCA in an undefined state.
995	 */
996	err = mthca_reset(mdev);
997	if (err) {
998		mthca_err(mdev, "Failed to reset HCA, aborting.\n");
999		goto err_free_dev;
1000	}
1001
1002	if (msi_x && !mthca_enable_msi_x(mdev))
1003		mdev->mthca_flags |= MTHCA_FLAG_MSI_X;
1004	if (msi && !(mdev->mthca_flags & MTHCA_FLAG_MSI_X) &&
1005	    !pci_enable_msi(pdev))
1006		mdev->mthca_flags |= MTHCA_FLAG_MSI;
1007
1008	if (mthca_cmd_init(mdev)) {
1009		mthca_err(mdev, "Failed to init command interface, aborting.\n");
1010		goto err_free_dev;
1011	}
1012
1013	err = mthca_tune_pci(mdev);
1014	if (err)
1015		goto err_cmd;
1016
1017	err = mthca_init_hca(mdev);
1018	if (err)
1019		goto err_cmd;
1020
1021	if (mdev->fw_ver < mthca_hca_table[id->driver_data].latest_fw) {
1022		mthca_warn(mdev, "HCA FW version %x.%x.%x is old (%x.%x.%x is current).\n",
1023			   (int) (mdev->fw_ver >> 32), (int) (mdev->fw_ver >> 16) & 0xffff,
1024			   (int) (mdev->fw_ver & 0xffff),
1025			   (int) (mthca_hca_table[id->driver_data].latest_fw >> 32),
1026			   (int) (mthca_hca_table[id->driver_data].latest_fw >> 16) & 0xffff,
1027			   (int) (mthca_hca_table[id->driver_data].latest_fw & 0xffff));
1028		mthca_warn(mdev, "If you have problems, try updating your HCA FW.\n");
1029	}
1030
1031	err = mthca_setup_hca(mdev);
1032	if (err)
1033		goto err_close;
1034
1035	err = mthca_register_device(mdev);
1036	if (err)
1037		goto err_cleanup;
1038
1039	err = mthca_create_agents(mdev);
1040	if (err)
1041		goto err_unregister;
1042
1043	pci_set_drvdata(pdev, mdev);
1044
1045	return 0;
1046
1047err_unregister:
1048	mthca_unregister_device(mdev);
1049
1050err_cleanup:
1051	mthca_cleanup_mcg_table(mdev);
1052	mthca_cleanup_av_table(mdev);
1053	mthca_cleanup_qp_table(mdev);
1054	mthca_cleanup_cq_table(mdev);
1055	mthca_cmd_use_polling(mdev);
1056	mthca_cleanup_eq_table(mdev);
1057
1058	mthca_pd_free(mdev, &mdev->driver_pd);
1059
1060	mthca_cleanup_mr_table(mdev);
1061	mthca_cleanup_pd_table(mdev);
1062	mthca_cleanup_uar_table(mdev);
1063
1064err_close:
1065	mthca_close_hca(mdev);
1066
1067err_cmd:
1068	mthca_cmd_cleanup(mdev);
1069
1070err_free_dev:
1071	if (mdev->mthca_flags & MTHCA_FLAG_MSI_X)
1072		pci_disable_msix(pdev);
1073	if (mdev->mthca_flags & MTHCA_FLAG_MSI)
1074		pci_disable_msi(pdev);
1075
1076	ib_dealloc_device(&mdev->ib_dev);
1077
1078err_free_res:
1079	mthca_release_regions(pdev, ddr_hidden);
1080
1081err_disable_pdev:
1082	pci_disable_device(pdev);
1083	pci_set_drvdata(pdev, NULL);
1084	return err;
1085}
1086
1087static void __devexit mthca_remove_one(struct pci_dev *pdev)
1088{
1089	struct mthca_dev *mdev = pci_get_drvdata(pdev);
1090	u8 status;
1091	int p;
1092
1093	if (mdev) {
1094		mthca_free_agents(mdev);
1095		mthca_unregister_device(mdev);
1096
1097		for (p = 1; p <= mdev->limits.num_ports; ++p)
1098			mthca_CLOSE_IB(mdev, p, &status);
1099
1100		mthca_cleanup_mcg_table(mdev);
1101		mthca_cleanup_av_table(mdev);
1102		mthca_cleanup_qp_table(mdev);
1103		mthca_cleanup_cq_table(mdev);
1104		mthca_cmd_use_polling(mdev);
1105		mthca_cleanup_eq_table(mdev);
1106
1107		mthca_pd_free(mdev, &mdev->driver_pd);
1108
1109		mthca_cleanup_mr_table(mdev);
1110		mthca_cleanup_pd_table(mdev);
1111
1112		iounmap(mdev->kar);
1113		mthca_uar_free(mdev, &mdev->driver_uar);
1114		mthca_cleanup_uar_table(mdev);
1115		mthca_close_hca(mdev);
1116		mthca_cmd_cleanup(mdev);
1117
1118		if (mdev->mthca_flags & MTHCA_FLAG_MSI_X)
1119			pci_disable_msix(pdev);
1120		if (mdev->mthca_flags & MTHCA_FLAG_MSI)
1121			pci_disable_msi(pdev);
1122
1123		ib_dealloc_device(&mdev->ib_dev);
1124		mthca_release_regions(pdev, mdev->mthca_flags &
1125				      MTHCA_FLAG_DDR_HIDDEN);
1126		pci_disable_device(pdev);
1127		pci_set_drvdata(pdev, NULL);
1128	}
1129}
1130
1131static struct pci_device_id mthca_pci_table[] = {
1132	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_TAVOR),
1133	  .driver_data = TAVOR },
1134	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_TAVOR),
1135	  .driver_data = TAVOR },
1136	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT),
1137	  .driver_data = ARBEL_COMPAT },
1138	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT),
1139	  .driver_data = ARBEL_COMPAT },
1140	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_ARBEL),
1141	  .driver_data = ARBEL_NATIVE },
1142	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_ARBEL),
1143	  .driver_data = ARBEL_NATIVE },
1144	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_SINAI),
1145	  .driver_data = SINAI },
1146	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_SINAI),
1147	  .driver_data = SINAI },
1148	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_SINAI_OLD),
1149	  .driver_data = SINAI },
1150	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_SINAI_OLD),
1151	  .driver_data = SINAI },
1152	{ 0, }
1153};
1154
1155MODULE_DEVICE_TABLE(pci, mthca_pci_table);
1156
1157static struct pci_driver mthca_driver = {
1158	.name		= DRV_NAME,
1159	.id_table	= mthca_pci_table,
1160	.probe		= mthca_init_one,
1161	.remove		= __devexit_p(mthca_remove_one)
1162};
1163
1164static int __init mthca_init(void)
1165{
1166	int ret;
1167
1168	ret = pci_register_driver(&mthca_driver);
1169	return ret < 0 ? ret : 0;
1170}
1171
1172static void __exit mthca_cleanup(void)
1173{
1174	pci_unregister_driver(&mthca_driver);
1175}
1176
1177module_init(mthca_init);
1178module_exit(mthca_cleanup);
1179